Credit Scoring Models Research Articles

Application of the VNS heuristic for feature selection in credit scoring problems

Credit scoring plays a major role for financial institutions when making credit-granting decisions. In this context, machine learning techniques have been used to develop credit scoring models, as they seek to recognize existing patterns in databases containing the credit history of borrowers to infer potential defaulters. However, these databases often contain a large number of variables, some of which can be noisy, leading to imprecise results and loss of performance/accuracy. In the present work, a feature selection technique is proposed based on a variable neighborhood concept, so-called VNS. The applicability of the method is assessed in conjunction with seven of the main techniques used to make default prediction in credit analysis problems. Its performance was compared to the feature selection obtained by the well-known PCA statistical method. The results indicate superior performance of the VNS in most of the applied tests, suggesting the robustness of the method.

A CWGAN-GP-based multi-task learning model for consumer credit scoring

In consumer credit scoring practice, there is often an imbalanced distribution in accepted borrowers, which means there are far fewer defaulters than borrowers who pay on time. This makes it difficult for traditional models to function. Aside from traditional sampling methods for imbalanced data, the idea of using rejected information to one’s benefit is new. Without historical repayment performance, rejected data are often discarded or simply disposed of during credit scoring modeling. However, these data play an important role because they capture the distribution of the borrower population as well as the accepted data. Besides, due to the increasing complexity in loan businesses, the current methods have difficulties in addressing high-dimensional multi-source data. Thus, a more effective credit scoring approach towards imbalanced data should be studied. Inspired by the state-of-the-art neural network methods, in this paper, we propose a conditional Wasserstein generative adversarial network with a gradient penalty (CWGAN-GP)-based multi-task learning (MTL) model (CWGAN-GP-MTL) for consumer credit scoring. First, the CWGAN-GP model is employed to learn about the distribution of the borrower population given both accepted and rejected data. Then, the data distribution between good and bad borrowers is adjusted through augmenting synthetic bad data generated by CWGAN-GP. Next, we design an MTL framework for both accepted and rejected and good and bad data, which improves risk prediction ability through parameter sharing. The proposed model was evaluated on real-world consumer loan datasets from a Chinese financial technology company. The empirical results indicate that the proposed model performed better than baseline models across different evaluation metrics, demonstrating its promising application potential.

Semi-supervised adapted HMMs for P2P credit scoring systems with reject inference.

The majority of current credit-scoring models, used for loan approval processing, are generally built on the basis of the information from the accepted credit applicants whose ability to repay the loan is known. This situation generates what is called the selection bias, presented by a sample that is not representative of the population of applicants, since rejected applications are excluded. Thus, the impact on the eligibility of those models from a statistical and economic point of view. Especially for the models used in the peer-to-peer lending platforms, since their rejection rate is extremely high. The method of inferring rejected applicants information in the process of construction of the credit scoring models is known as reject inference. This study proposes a semi-supervised learning framework based on hidden Markov models (SSHMM), as a novel method of reject inference. Real data from the Lending Club platform, the most used online lending marketplace in the United States as well as the rest of the world, is used to experiment the effectiveness of our method over existing approaches. The results of this study clearly illustrate the proposed method’s superiority, stability, and adaptability.

A credit scoring model based on the Myers–Briggs type indicator in online peer-to-peer lending

Although psychometric features have been considered for alternative credit scoring, they have not yet been applied to peer-to-peer (P2P) lending because such information is not available on platforms. This study proposed an alternative credit scoring model for P2P lending by extracting typical personality types inferred from the borrowers’ job category. We projected a virtual space of borrowers by using the affinity matrix based on the Myers–Briggs type indicator (MBTI) that fits each job category. Applying the distance in this space to Lending Club data, we used locally weighted logistic regression to vary the coefficients of the variables, which affect loan repayments, with each MBTI type for predicting the default probability. We found that each MBTI type’s credit scoring model has different significant variables. This study provides insights into breakthroughs in developing alternative credit scoring for P2P lending.

An uncertainty-oriented cost-sensitive credit scoring framework with multi-objective feature selection

• An uncertainty-oriented credit scoring framework with multi-objective feature selection is developed to tackle the credit classification task under uncertainty. • This study extends the use of cost space to feature selection in the credit scoring process. • The proposed credit scoring framework can provide a series of Pareto-optimal credit scoring models to fit different decision-making contexts. • The cost plot enables credit decision-makers to see the results of different credit scoring models with different risk sources and corresponding expected costs. In order to solve the problem of uncertain misclassification costs and class distributions in credit scoring tasks, an uncertainty-oriented credit scoring framework based on a multi-objective feature selection strategy is proposed in this study. This proposed framework searches for a pool of Pareto-optimal credit scoring models with different feature subsets without the assumption of the operating condition (misclassification costs and class distributions) information. Specifically, the searching process concerns the trade-off of the False Positive Rate and the False Negative Rate using a binary multi-objective particle swarm optimization (BMOPSO) algorithm. By visualizing the Pareto-optimal solutions in cost space, credit decision-makers can select an optimal compromise model based on their decision-making contexts. The proposed framework is compared with baseline models on three retail credit scoring datasets. The experimental results show that the proposed framework could find out the optimal credit model with minimal misclassification cost for almost all possible operating conditions.

Machine learning interpretability for a stress scenario generation in credit scoring based on counterfactuals

To boost the application of machine learning (ML) techniques for credit scoring models, the blackbox problem should be addressed. The primary aim of this paper is to propose a measure based on counterfactuals to evaluate the interpretability of a ML credit scoring technique. Counterfactuals assist with understanding the model with regard to the classification decision boundaries and evaluate model robustness. The second contribution is the development of a data perturbation technique to generate a stress scenario.We apply these two proposals to a dataset on UK unsecured personal loans to compare logistic regression and stochastic gradient boosting (SBG). We show that training a blackbox model (SGB) as conditioned on our data perturbation technique can provide insight into model performance under stressed scenarios. The empirical results show that our interpretability measure is able to capture the classification decision boundary, unlike AUC and the classification accuracy widely used in the banking sector.

Default or profit scoring credit systems? Evidence from European and US peer-to-peer lending markets

For the emerging peer-to-peer (P2P) lending markets to survive, they need to employ credit-risk management practices such that an investor base is profitable in the long run. Traditionally, credit-risk management relies on credit scoring that predicts loans’ probability of default. In this paper, we use a profit scoring approach that is based on modeling the annualized adjusted internal rate of returns of loans. To validate our profit scoring models with traditional credit scoring models, we use data from a European P2P lending market, Bondora, and also a random sample of loans from the Lending Club P2P lending market. We compare the out-of-sample accuracy and profitability of the credit and profit scoring models within several classes of statistical and machine learning models including the following: logistic and linear regression, lasso, ridge, elastic net, random forest, and neural networks. We found that our approach outperforms standard credit scoring models for Lending Club and Bondora loans. More specifically, as opposed to credit scoring models, returns across all loans are 24.0% (Bondora) and 15.5% (Lending Club) higher, whereas accuracy is 6.7% (Bondora) and 3.1% (Lending Club) higher for the proposed profit scoring models. Moreover, our results are not driven by manual selection as profit scoring models suggest investing in more loans. Finally, even if we consider data sampling bias, we found that the set of superior models consists almost exclusively of profit scoring models. Thus, our results contribute to the literature by suggesting a paradigm shift in modeling credit-risk in the P2P market to prefer profit as opposed to credit-risk scoring models.

Deep reinforcement learning with the confusion-matrix-based dynamic reward function for customer credit scoring

Customer credit scoring is a dynamic interactive process. Simply designing the static reward function for deep reinforcement learning may be difficult to guide an agent to adapt to the change of the customer credit scoring environment. To solve this problem, we propose the deep Q-network with the confusion-matrix-based dynamic reward function (DQN-CMDRF) model. Especially, the new constructed dynamic reward function can adjust the reward dynamically according to the change of confusion matrix after each deep Q-network model training, which can guide the agent to adapt to the change of environment quickly, so as to improve the customer credit scoring performance of the deep Q-network model. First, we formulate customer credit scoring as a finite Markov decision process. Second, to adjust the reward dynamically according to the customer credit scoring environment, the dynamic reward function is designed based on the confusion matrix. Finally, we introduce the confusion-matrix-based dynamic reward function into the deep Q-network model for customer credit scoring. To verify the effectiveness of the proposed model, we introduce four evaluation measures and make a series of experiments on the five customer credit scoring datasets. The experimental results show that the constructed dynamic reward function can more effectively improve customer credit scoring performance of the deep Q-network model, and the performance of the DQN-CMDRF model is significantly better than that of the other eight traditional classification models. More importantly, we find that the constructed dynamic reward function can accelerate the convergence speed and improve the stability of the deep Q-network model.

A Monte Carlo simulation framework for reject inference

Credit scoring is the process of determining whether applicants should be granted a financial loan. When a financial institution decides to create a credit scoring model for all applicants, the institution only has the known good/bad loan outcomes for accepted applicants. This causes inherent bias in the model We address a gap in the reject inference literature by developing a methodology to simulate rejected applicants. A methodology to illustrate how to simulate rejected applicants must be developed so that the reject inference techniques can be studied and appropriate reject inference techniques can be selected. This study uses a peer-to-peer financial loan information from accepted and rejected financial loan applicants to perform Monte Carlo simulation of rejected applicants. Using simulated data, the researchers compare the performance of three widely used reject inference techniques.

Efficient multi-classifier wrapper feature-selection model: Application for dimension reduction in credit scoring

The task of identifying most relevant features for a credit scoring application is a challenging task. Reducing the number of redundant and unwanted features is an inevitable task to improve the performance of the credit scoring model. The wrappers approach is usually used in credit scoring applications to identify the most relevant features. However, this approach suffers from the issue of subsets generation and the use of a single classifier as an evaluation function. The problem here is that each classifier may give different results which can be interpreted differently. Hence, we propose in this study an ensemble wrapper feature selection model which is based on a multi-classifiers combination. In a first stage, we address the problem of subsets generation by minimizing the search space through a customized heuristic. Then, a multi-classifier wrapper evaluation is applied using two classifier arrangement approaches in order to select a set of mutually approved set of relevant features. The proposed method is evaluated on four credit datasets and has shown a good performance compared to individual classifiers results.

Novel hybrid ensemble credit scoring model with stacking-based noise detection and weight assignment

Credit scoring is used to help financial institutions control default risks and reduce economic losses, and a variety of mainstream machine learning and data mining algorithms have been applied for this purpose. However, real-world datasets are generally noisy, which seriously affects the performance of credit scoring models. Among the mainstream strategies for handling noise, instance filtering may result in information loss, especially for hard-to-access credit datasets, and label noise correction may produce erroneous information in the dataset. In this study, to reduce the adverse impact of noisy data on the performance of classification algorithms, a novel hybrid ensemble credit scoring model with stacking-based noise detection and weight assignment is developed to remove or adapt noisy data in raw datasets and to form noise-detected training data to obtain excellent default risk prediction competence. Furthermore, a new weight assignment approach based on a cloud model is proposed, which is applied to calculate the weight values of the classifiers in the weighted voting ensemble model to improve the prediction accuracy of the proposed model. In this study, five public datasets are adopted using five performance metrics to evaluate the performance of the proposed model. The experimental results demonstrate good model prediction power and robustness.

Generating customer’s credit behavior with deep generative models

Banks collect data x1 in loan applications to decide whether to grant credit and accepted applications generate new data x2 throughout the loan period. Hence, banks have two measurement-modalities, which provide a complete picture about customers. If we can generate x2 conditioned on x1 keeping the relationship between these two modalities, credit and behavior scoring may be enabled simultaneously (at the time x1 is obtained) to support cross-selling, launching of new products or marketing campaigns. Therefore, we develop a novel conditional bi-modal discriminative (CBMD) model for credit scoring, which is able to generate x2 based on x1 and can classify the outcome of loans in an unified framework. The idea behind CBMD is to learn joint (among modalities) latent representations that are useful to generate x2 using the available data x1 during the application process. The classifier model introduced in CBMD encourages the generative process to generate x2 accurately. Further, CBMD optimizes a novel objective function introduced in this research, which maximizes mutual information between the latent representation z and the modality x2 to improve the generative process in the model. We benchmark the generative process of our proposed model and CBMD outperforms other multi-learning models. Similarly, the classification performance of CBMD is tested under different scenarios and it achieves higher or on a par model performance compared to the state-of-the-art in multi-modal learning models.

Prediction Model for Loan Default Using Machine Learning

Lack of appropriate predictive models for prompt loan assessment in granting loan application leads most financial lending institutions to bankruptcy. Loan default predictions models are used as critical tools in loan application assessment to discriminate between ‘bad' and ‘good' loan applicants. Recently, classification and survival techniques have been applied in loan default prediction with revolving cases of default over prediction. Despite advancements in automating decision-based loan systems, most existing models do not consider the ‘early loan repayment' attribute as a factor in resolving this prediction error. In practice, the adjustment for early loan repayment in model building is necessary because a high rate of early loan repayment decreases the number of defaults observed in a portfolio. To achieve this, three supervised machine learning algorithms i.e. logistic regression, Support vector machine and Naí¯ve B were used to develop a credit scoring models which include the early loan repayment attribute. The models were trained and tested on a loan dataset consisting of attributes with, and without early loan repayment attribute. Models evaluations were done using five performance metrics. The results of the performance evaluation shows that models that account for early loan repayment have higher accuracy, recall, precision, RMSE and ROC values than models trained without the early loan repayment attribute. The Logistic regression algorithm outperformed all other algorithms with 91% accuracy, 11% RMSE, 89% precision and 88% recall values.

Promoting Variable Effect Consistency in Mixture Cure Model for Credit Scoring

Mixture cure models are widely adopted in credit scoring. Mixture cure models consist of two parts: an incident part which predicts the probability of default and a latency part which predicts when they are likely to default. The two model parts describe two quite relevant credit aspects. So, it is reasonable to expect that the two sets of the coefficients are somewhat related. Moreover, in practical cases, it is difficult to interpret the results when the two sets of the coefficients of the same variables have conflicting signs. Most existing works either ignore the interconnections of the two sets of coefficients or impose a strict constraint between them. We proposed a mixture cure model considering the variable effect consistency using a sign-based penalty. It is a more flexible model that allows the two sets of coefficients to be in different distributions and magnitudes. To accommodate high-dimensional credit data, a group lasso penalty is also imposed for variable selection. Simulation shows that the proposed method has competitive performance compared with alternative methods in terms of estimation and prediction. Furthermore, the empirical study illustrates that the proposed method outperforms the alternative method and can improve the interpretability of the results.

A novel multi-stage ensemble model for credit scoring based on synthetic sampling and feature transformation

Credit scoring has become increasingly important for financial institutions. With the advancement of artificial intelligence, machine learning methods, especially ensemble learning methods, have become increasingly popular for credit scoring. However, the problems of imbalanced data distribution and underutilized feature information have not been well addressed sufficiently. To make the credit scoring model more adaptable to imbalanced datasets, the original model-based synthetic sampling method is extended herein to balance the datasets by generating appropriate minority samples to alleviate class overlap. To enable the credit scoring model to extract inherent correlations from features, a new bagging-based feature transformation method is proposed, which transforms features using a tree-based algorithm and selects features using the chi-square statistic. Furthermore, a two-layer ensemble method that combines the advantages of dynamic ensemble selection and stacking is proposed to improve the classification performance of the proposed multi-stage ensemble model. Finally, four standardized datasets are used to evaluate the performance of the proposed ensemble model using six evaluation metrics. The experimental results confirm that the proposed ensemble model is effective in improving classification performance and is superior to other benchmark models.

A hybrid model with novel feature selection method and enhanced voting method for credit scoring

With the in-depth application of artificial intelligence technology in the financial field, credit scoring models constructed by machine learning algorithms have become mainstream. However, the high-dimensional and complex attribute features of the borrower pose challenges to the predictive competence of the model. This paper proposes a hybrid model with a novel feature selection method and an enhanced voting method for credit scoring. First, a novel feature selection combined method based on a genetic algorithm (FSCM-GA) is proposed, in which different classifiers are used to select features in combination with a genetic algorithm and combine them to generate an optimal feature subset. Furthermore, an enhanced voting method (EVM) is proposed to integrate classifiers, with the aim of improving the classification results in which the prediction probability values are close to the threshold. Finally, the predictive competence of the proposed model was validated on three public datasets and five evaluation metrics (accuracy, AUC, F-score, Log loss and Brier score). The comparative experiment and significance test results confirmed the good performance and robustness of the proposed model.

An Optimal Credit Scoring Model Based on the Maximum Default Identification Ability for Chinese Small Business

The reasonable credit scoring model must have strong default identification ability, which means the credit scoring can effectively distinguish between defaulting and nondefaulting customers. The premise to determine the credit score of small enterprises is to determine the weight of indicators. This paper studies 3,045 Chinese small business loans, and two novel weighting methods “Wilks’ Lambda method” and “AUC value method” are proposed, The greater the weight they meet, the greater the ability of default identification. The five weighting methods of “Wilks’ lambda method,” “AUC value method,” “G1 method,” “entropy method,” and “mean square variance method” are compared. An important contribution of the paper is to discover that Wilks’ Lambda method is the most effective method for small business.

A deep learning model for behavioural credit scoring in banks

A deep learning model for behavioural credit scoring in banks

Machine learning and artificial neural networks to construct P2P lending credit-scoring model: A case using Lending Club data

<abstract> <p>In this study, we constructed the credit-scoring model of P2P loans by using several machine learning and artificial neural network (ANN) methods, including logistic regression (LR), a support vector machine, a decision tree, random forest, XGBoost, LightGBM and 2-layer neural networks. This study explores several hyperparameter settings for each method by performing a grid search and cross-validation to get the most suitable credit-scoring model in terms of training time and test performance. In this study, we get and clean the open P2P loan data from Lending Club with feature engineering concepts. In order to find significant default factors, we used an XGBoost method to pre-train all data and get the feature importance. The 16 selected features can provide economic implications for research about default prediction in P2P loans. Besides, the empirical result shows that gradient-boosting decision tree methods, including XGBoost and LightGBM, outperform ANN and LR methods, which are commonly used for traditional credit scoring. Among all of the methods, XGBoost performed the best.</p> </abstract>

An end-to-end deep learning approach to credit scoring using CNN C XGBoost on transaction data

The performance of credit scoring models is closely linked to a bank’s profitability. Application scoring models for potential customers usually perform worse than models for existing customers. This is due to the lender not having access to the financial behavioral data of potential customers. Access to such data about potential customers could therefore increase a bank’s profitability. Open banking application programming interfaces (APIs) provide access to 90 days of historical data on potential customers’ balances and transactions. We examine the performance of credit scoring models developed using such data from a Norwegian bank. We find that traditional regression models perform poorly, while machine learning (ML) methods can provide models that perform satisfactorily based on these data alone. Further, we find that the best performing models are based on an end-to-end deep learning approach, where machine learning algorithms create explanatory variables based on non-aggregated data. These results indicate that data available through the open banking APIs can be an important data source when banks assess the creditworthiness of potential customers. In combination with end-to-end deep learning methods they have the potential to increase the performance of a bank’s application credit scoring models and thus increase the bank’s profitability.